dari.dev
dari.dev is the routing layer for AI agents. It provides a hosted model router and open-source/self-hostable gateway and router framework so developers can run agents through one endpoint, select models based on workload and policy, and reduce token cost without changing their agent harness.
At a Glance
- Developers and teams building AI applications
- Engineering teams shipping coding agents
- B2B developer-tools and AI-infrastructure customers
- Enterprise teams needing private routing, workload-specific evaluation, and cost controls
AI Tools by dari.dev
(1)Dari Docs
CLI That Tests Docs With Agents
Discussions
No discussions yet
Be the first to start a discussion about dari.dev
Latest News
Dari CLI v1.0.281 released with selected thinking-level propagation to explicit models across harnesses.
Dari Docs documentation repository update restyled the README and workflow materials around the self-bench presentation.
Dari announced its versioned, stateful agent deployment API for long-running agents, with durable sessions, sandbox workspaces, retries, logs, credentials, and multimodal streaming.
Dari's public product direction centered on model routing, a hosted gateway, BYOK and subscription pooling, and integrations for coding-agent harnesses.
Products & Services
Hosted OpenAI-compatible routing for applications and coding agents. It chooses a model and thinking level for each request using routing policy, benchmark signals, context, and provider cache state; the managed platform provides configured models, credentials, retries, telemetry, persistent routing state, and speculative routing.
Self-hostable runtime behind Dari's hosted routing service. It exposes Anthropic Messages, OpenAI Chat Completions, and OpenAI Responses through one endpoint, supports provider API keys and connected Claude Code/ChatGPT subscriptions, and can run without a Dari account.
Open-source TypeScript framework for building and self-hosting routers with pluggable policies. It handles normalization, model eligibility, provider execution, failover, and protocol translation while the operator supplies models, credentials, and infrastructure.
Go CLI that authenticates users, manages routers and API keys, launches Claude Code, Codex, OpenCode, Pi, OpenClaw, and Hermes through Dari, and supports gateway and hosted-router workflows.
Market Position
Dari positions itself as a drop-in routing layer for AI agents rather than a single-model provider: it combines workload-aware model selection, cache awareness, subscription pooling, protocol normalization, and a self-hostable gateway. Its competitive set includes model gateways and routing platforms such as LiteLLM, OpenRouter, Portkey, and Braintrust-style evaluation/routing tooling; Dari differentiates around coding-agent harness integration, durable gateway/runtime behavior, and a trained routing policy/open-source stack.
Leadership
Founders
Avyay Varadarajan
Founder/CEO; studied computer science at Caltech and previously worked as a software engineer at Uber, in machine learning with the Lakers, and at startups.
Benjamin Hong
Founder/CTO; Caltech computer-science graduate and former software engineer at Amazon and Uber.
Executive Team
Avyay Varadarajan
Founder and CEO
Caltech computer-science background; previously software engineering at Uber, machine learning with the Lakers, and startups.
Benjamin Hong
Founder and CTO
Caltech computer-science background; previously worked as a software engineer at Amazon and Uber.
Founding Story
The founders started by building browser-automation agents and repeatedly encountered the need for a durable control plane and reliable infrastructure beneath production agents. They initially built Dari for themselves, then broadened the idea after seeing coding, research, and operations teams rebuilding durable execution, sandboxes, state, secrets, retries, and observability; the company subsequently focused its public product on model routing and a gateway for AI agents.
Business Model
Revenue Model
A freemium hosted gateway with usage-based managed-provider billing and paid autorouting subscriptions; team and enterprise plans add private routers, volume pricing, dedicated support, and potentially private router training. Self-hosted gateway users bring their own keys/subscriptions and infrastructure.
Pricing Tiers
Use personal AI subscriptions, Dari managed billing, or your own provider keys on one-model routers; no autorouting. Managed usage is billed as used.
Includes Gateway and routers with many models; Dari's routing router selects the model and thinking level for each request.
Private router, volume pricing, dedicated support, and enterprise options to train a private router on the team's workload and evaluations.
Target Markets
- Developers and teams building AI applications
- Engineering teams shipping coding agents
- B2B developer-tools and AI-infrastructure customers
- Enterprise teams needing private routing, workload-specific evaluation, and cost controls
- Running coding agents through an automatically selected model
- Reducing model/token costs while preserving quality on real coding repositories
- Routing production AI applications through a single API endpoint
- Using personal AI subscriptions or provider API keys across agents
- Self-hosting a model gateway with custom models, policies, and providers
- Training or configuring private routers on enterprise workload traces and evaluations